![Page 1: Metropolis methods applied to Bayesian geologic inversionsMetropolis methods applied to Bayesian geologic inversions Michael Glinsky (CEO Science Leader, CSIRO; Adjunct Professor of](https://reader033.vdocument.in/reader033/viewer/2022060209/5f04411f7e708231d40d11de/html5/thumbnails/1.jpg)
Metropolis methods applied to Bayesian geologic inversions
Michael Glinsky (CEO Science Leader, CSIRO; Adjunct Professor of Physics, UWA)
“Three Sisters” -- aboriginal womans’ place for doing business, near BHPB Yandi iron ore mine1Wednesday, 22 June 2011
![Page 2: Metropolis methods applied to Bayesian geologic inversionsMetropolis methods applied to Bayesian geologic inversions Michael Glinsky (CEO Science Leader, CSIRO; Adjunct Professor of](https://reader033.vdocument.in/reader033/viewer/2022060209/5f04411f7e708231d40d11de/html5/thumbnails/2.jpg)
Outline
• Bayesian inversion with uncertainty as a statistical mechanics problem
• Application to seismic inversion• Application to marine E&M inversion• The future
2Wednesday, 22 June 2011
![Page 3: Metropolis methods applied to Bayesian geologic inversionsMetropolis methods applied to Bayesian geologic inversions Michael Glinsky (CEO Science Leader, CSIRO; Adjunct Professor of](https://reader033.vdocument.in/reader033/viewer/2022060209/5f04411f7e708231d40d11de/html5/thumbnails/3.jpg)
The connection of Bayesian inversion to the Metropolis method of statistical mechanics
P(A^B) = P(A) P(B|A) = P(B) P(A|B)Bayes’ Theorem -- a probability commutation relation
A = m = modelB = d = observed data
P(m|d) = P(m) P(d|m) / P(d)want to know
know forward model
prior probability of model
log[ P(d|m) ] ~ - [ d - s(m) ]^2 / noise^2 ~ - H / kBT
m = mP(m | d)dm =
1N
mi for {mi }P (m|d )i=1
N
∑∫3Wednesday, 22 June 2011
![Page 4: Metropolis methods applied to Bayesian geologic inversionsMetropolis methods applied to Bayesian geologic inversions Michael Glinsky (CEO Science Leader, CSIRO; Adjunct Professor of](https://reader033.vdocument.in/reader033/viewer/2022060209/5f04411f7e708231d40d11de/html5/thumbnails/4.jpg)
CSIRO. CSIRO. Integration of uncertain subsurface information into multiple reservoir simulation models
Probabilistic model based inversion
• Layer based model built at seismic loop scale using sparse spike inversion
• Standard rock physics correlations estimated with uncertainty• Fundamental properties of layers are:
• net-to-gross ratio (N/G)• layer top and base• fluid type
• Ensemble of models generated that are consistent with seismic to within estimated noise level
4Wednesday, 22 June 2011
![Page 5: Metropolis methods applied to Bayesian geologic inversionsMetropolis methods applied to Bayesian geologic inversions Michael Glinsky (CEO Science Leader, CSIRO; Adjunct Professor of](https://reader033.vdocument.in/reader033/viewer/2022060209/5f04411f7e708231d40d11de/html5/thumbnails/5.jpg)
A closer look at Bayesian seismic inversion
• Fundamental parameters– Layer times– Rock properties in each layer– Fluid type
• Forward model– Reuss/Gassman for fluids– Convolution (multi-stack)
• Priors– Regional rock trends, layer “picks”
• Likelihoods– Synthetic seismic– Isopachs
• Posteriors– Multimodel MCMC sampling
5Wednesday, 22 June 2011
![Page 6: Metropolis methods applied to Bayesian geologic inversionsMetropolis methods applied to Bayesian geologic inversions Michael Glinsky (CEO Science Leader, CSIRO; Adjunct Professor of](https://reader033.vdocument.in/reader033/viewer/2022060209/5f04411f7e708231d40d11de/html5/thumbnails/6.jpg)
Ensemble of models at well location show effect of model based inversion
soft
hard
acoustic impedance
before after
realization realization
2.2 s
2.4 s
6Wednesday, 22 June 2011
![Page 7: Metropolis methods applied to Bayesian geologic inversionsMetropolis methods applied to Bayesian geologic inversions Michael Glinsky (CEO Science Leader, CSIRO; Adjunct Professor of](https://reader033.vdocument.in/reader033/viewer/2022060209/5f04411f7e708231d40d11de/html5/thumbnails/7.jpg)
Effect of model based inversion on match of synthetic seismic to seismic data
2.24 s
2.36 s
before after seismic
7Wednesday, 22 June 2011
![Page 8: Metropolis methods applied to Bayesian geologic inversionsMetropolis methods applied to Bayesian geologic inversions Michael Glinsky (CEO Science Leader, CSIRO; Adjunct Professor of](https://reader033.vdocument.in/reader033/viewer/2022060209/5f04411f7e708231d40d11de/html5/thumbnails/8.jpg)
Inversion tightens the range of possible net sand
before
after
well value0 m 50 m
probability of oil increased to 97% from 50% (oil in sand at this location)
8Wednesday, 22 June 2011
![Page 9: Metropolis methods applied to Bayesian geologic inversionsMetropolis methods applied to Bayesian geologic inversions Michael Glinsky (CEO Science Leader, CSIRO; Adjunct Professor of](https://reader033.vdocument.in/reader033/viewer/2022060209/5f04411f7e708231d40d11de/html5/thumbnails/9.jpg)
Imbed the result into 3D model
9Wednesday, 22 June 2011
![Page 10: Metropolis methods applied to Bayesian geologic inversionsMetropolis methods applied to Bayesian geologic inversions Michael Glinsky (CEO Science Leader, CSIRO; Adjunct Professor of](https://reader033.vdocument.in/reader033/viewer/2022060209/5f04411f7e708231d40d11de/html5/thumbnails/10.jpg)
An overview of the model based inversion process
10Wednesday, 22 June 2011
![Page 11: Metropolis methods applied to Bayesian geologic inversionsMetropolis methods applied to Bayesian geologic inversions Michael Glinsky (CEO Science Leader, CSIRO; Adjunct Professor of](https://reader033.vdocument.in/reader033/viewer/2022060209/5f04411f7e708231d40d11de/html5/thumbnails/11.jpg)
Another application of Metropolis method -- marine CSEM inversion
• description of the physics• what makes the inversion difficult
• tight & non-linear (parametric bootstrap & path finding)• multi modal (model selection)• bound constraints (projected newton methods)
• connection of Bayesian to classical methods• flavors of Bayes
• Bayesian smoothing • layer split/merge• log grid with no-smoothing
• benchmarks• wedge• “bird” model
• anomaly definition• systematic noise
11Wednesday, 22 June 2011
![Page 12: Metropolis methods applied to Bayesian geologic inversionsMetropolis methods applied to Bayesian geologic inversions Michael Glinsky (CEO Science Leader, CSIRO; Adjunct Professor of](https://reader033.vdocument.in/reader033/viewer/2022060209/5f04411f7e708231d40d11de/html5/thumbnails/12.jpg)
Appeal? A “hydrocarbon saturation” “reservoir quality” detector.
Very rapid growth of service companies since 2000. Psuedo-”bust” 2007-8.
Abundant data. Little consensus on interpretation.
Controlled Source ElectroMagnetics (CSEM) or seabed logging (SBL)
12Wednesday, 22 June 2011
![Page 13: Metropolis methods applied to Bayesian geologic inversionsMetropolis methods applied to Bayesian geologic inversions Michael Glinsky (CEO Science Leader, CSIRO; Adjunct Professor of](https://reader033.vdocument.in/reader033/viewer/2022060209/5f04411f7e708231d40d11de/html5/thumbnails/13.jpg)
Typical data and a 1D layered earth model fit
13Wednesday, 22 June 2011
![Page 14: Metropolis methods applied to Bayesian geologic inversionsMetropolis methods applied to Bayesian geologic inversions Michael Glinsky (CEO Science Leader, CSIRO; Adjunct Professor of](https://reader033.vdocument.in/reader033/viewer/2022060209/5f04411f7e708231d40d11de/html5/thumbnails/14.jpg)
need to Bayesian model average for y− P(y|D) = ∑ k P(y|Mk,D) P(D|k)
because of mixture distributions− model geometry (e.g., number of layers)− rock types− fluids
important because many times− uncertainties within a model < uncertainty between models
Multi-modal distributions -- an inversion difficulty
y
P(y|D)
14Wednesday, 22 June 2011
![Page 15: Metropolis methods applied to Bayesian geologic inversionsMetropolis methods applied to Bayesian geologic inversions Michael Glinsky (CEO Science Leader, CSIRO; Adjunct Professor of](https://reader033.vdocument.in/reader033/viewer/2022060209/5f04411f7e708231d40d11de/html5/thumbnails/15.jpg)
Model selection
x
y
Model selection – classic statistics problem
line ??, parabola ??, sextic ??
Newton or Ptolemy?
There exists sophisticated Bayesian model-selection procedures for general nonlinear regression problems.
15Wednesday, 22 June 2011
![Page 16: Metropolis methods applied to Bayesian geologic inversionsMetropolis methods applied to Bayesian geologic inversions Michael Glinsky (CEO Science Leader, CSIRO; Adjunct Professor of](https://reader033.vdocument.in/reader033/viewer/2022060209/5f04411f7e708231d40d11de/html5/thumbnails/16.jpg)
Tight & non-linear -- an inversion difficulty
m1
m2
Null space of F
seismic case
m1
m2
CSEM caseNull space of F
no Born approximation because of strong contrasts; therefore multimodal, badly scaled, and contorted in shape
16Wednesday, 22 June 2011
![Page 17: Metropolis methods applied to Bayesian geologic inversionsMetropolis methods applied to Bayesian geologic inversions Michael Glinsky (CEO Science Leader, CSIRO; Adjunct Professor of](https://reader033.vdocument.in/reader033/viewer/2022060209/5f04411f7e708231d40d11de/html5/thumbnails/17.jpg)
The problem and the solution
m1
m2
m1
m2
????
1. Parametric bootstrapmultiple start optimizationperturb data with noise
2. Change coordinates to mode connection paths
problem solutions
17Wednesday, 22 June 2011
![Page 18: Metropolis methods applied to Bayesian geologic inversionsMetropolis methods applied to Bayesian geologic inversions Michael Glinsky (CEO Science Leader, CSIRO; Adjunct Professor of](https://reader033.vdocument.in/reader033/viewer/2022060209/5f04411f7e708231d40d11de/html5/thumbnails/18.jpg)
A
B
Mj
Mi
m2
m3
(i=node index along path)
!2
evolution
evolution
0.5 1.0 1.5 2.0 2.50.5
1.0
1.5
2.0
2.5
0 5 10 15 20 25 30 350
1000
2000
3000
4000
5000
6000
7000
Mode connection paths
18Wednesday, 22 June 2011
![Page 19: Metropolis methods applied to Bayesian geologic inversionsMetropolis methods applied to Bayesian geologic inversions Michael Glinsky (CEO Science Leader, CSIRO; Adjunct Professor of](https://reader033.vdocument.in/reader033/viewer/2022060209/5f04411f7e708231d40d11de/html5/thumbnails/19.jpg)
An more realistic example of mode connection paths
!2E!1 2E!1 6E!1 1E0 1.4E0!2E!1
0E0
2E!1
4E!1
6E!1
8E!1
1E0
1.2E0
1.4E0 m_5_vs_m_4
!2E!1 2E!1 6E!1 1E0 1.4E0!2E!1
0E0
2E!1
4E!1
6E!1
8E!1
1E0
1.2E0
1.4E0 m_6_vs_m_4
!2E!1 2E!1 6E!1 1E0 1.4E0!2E!1
0E0
2E!1
4E!1
6E!1
8E!1
1E0
1.2E0
1.4E0 m_6_vs_m_5
!2E!1 2E!1 6E!1 1E0 1.4E0!2E!1
0E0
2E!1
4E!1
6E!1
8E!1
1E0
1.2E0
1.4E0 m_12_vs_m_4
!2E!1 2E!1 6E!1 1E0 1.4E0!2E!1
0E0
2E!1
4E!1
6E!1
8E!1
1E0
1.2E0
1.4E0 m_12_vs_m_5
!2E!1 2E!1 6E!1 1E0 1.4E0!2E!1
0E0
2E!1
4E!1
6E!1
8E!1
1E0
1.2E0
1.4E0
m_12_vs_m_6
!2E!1 2E!1 6E!1 1E0 1.4E0!2E!1
0E0
2E!1
4E!1
6E!1
8E!1
1E0
1.2E0
1.4E0 m_13_vs_m_4
!2E!1 2E!1 6E!1 1E0 1.4E0!2E!1
0E0
2E!1
4E!1
6E!1
8E!1
1E0
1.2E0
1.4E0 m_13_vs_m_5
!2E!1 2E!1 6E!1 1E0 1.4E0!2E!1
0E0
2E!1
4E!1
6E!1
8E!1
1E0
1.2E0
1.4E0 m_13_vs_m_6
!2E!1 2E!1 6E!1 1E0 1.4E0!2E!1
0E0
2E!1
4E!1
6E!1
8E!1
1E0
1.2E0
1.4E0 m_13_vs_m_12
!2E!1 2E!1 6E!1 1E0 1.4E0!2E!1
0E0
2E!1
4E!1
6E!1
8E!1
1E0
1.2E0
1.4E0 m_14_vs_m_4
!2E!1 2E!1 6E!1 1E0 1.4E0!2E!1
0E0
2E!1
4E!1
6E!1
8E!1
1E0
1.2E0
1.4E0 m_14_vs_m_5
!2E!1 2E!1 6E!1 1E0 1.4E0!2E!1
0E0
2E!1
4E!1
6E!1
8E!1
1E0
1.2E0
1.4E0 m_14_vs_m_6
!2E!1 2E!1 6E!1 1E0 1.4E0!2E!1
0E0
2E!1
4E!1
6E!1
8E!1
1E0
1.2E0
1.4E0 m_14_vs_m_12
!2E!1 2E!1 6E!1 1E0 1.4E0!2E!1
0E0
2E!1
4E!1
6E!1
8E!1
1E0
1.2E0
1.4E0 m_14_vs_m_13
Node index i=0,1..31
mode A mode B
4,5,6
1314
12
Dep
th(m
)
19Wednesday, 22 June 2011
![Page 20: Metropolis methods applied to Bayesian geologic inversionsMetropolis methods applied to Bayesian geologic inversions Michael Glinsky (CEO Science Leader, CSIRO; Adjunct Professor of](https://reader033.vdocument.in/reader033/viewer/2022060209/5f04411f7e708231d40d11de/html5/thumbnails/20.jpg)
Truncated distributions -- an inversion difficulty
20Wednesday, 22 June 2011
![Page 21: Metropolis methods applied to Bayesian geologic inversionsMetropolis methods applied to Bayesian geologic inversions Michael Glinsky (CEO Science Leader, CSIRO; Adjunct Professor of](https://reader033.vdocument.in/reader033/viewer/2022060209/5f04411f7e708231d40d11de/html5/thumbnails/21.jpg)
Connections to geostatistics
21Wednesday, 22 June 2011
![Page 22: Metropolis methods applied to Bayesian geologic inversionsMetropolis methods applied to Bayesian geologic inversions Michael Glinsky (CEO Science Leader, CSIRO; Adjunct Professor of](https://reader033.vdocument.in/reader033/viewer/2022060209/5f04411f7e708231d40d11de/html5/thumbnails/22.jpg)
Bayesian smoothing formulation
µ
2 χ2
(1)
(2)
(1) & (2) combinedoptimum
New piece
22Wednesday, 22 June 2011
![Page 23: Metropolis methods applied to Bayesian geologic inversionsMetropolis methods applied to Bayesian geologic inversions Michael Glinsky (CEO Science Leader, CSIRO; Adjunct Professor of](https://reader033.vdocument.in/reader033/viewer/2022060209/5f04411f7e708231d40d11de/html5/thumbnails/23.jpg)
Two flavors of Bayesian inversion
Bayesian SmoothingBayesian model-selection (splitting/merging)
23Wednesday, 22 June 2011
![Page 24: Metropolis methods applied to Bayesian geologic inversionsMetropolis methods applied to Bayesian geologic inversions Michael Glinsky (CEO Science Leader, CSIRO; Adjunct Professor of](https://reader033.vdocument.in/reader033/viewer/2022060209/5f04411f7e708231d40d11de/html5/thumbnails/24.jpg)
Comparison of the two flavors
Bayesian smoothing
Bayesian splittingmodel
24Wednesday, 22 June 2011
![Page 25: Metropolis methods applied to Bayesian geologic inversionsMetropolis methods applied to Bayesian geologic inversions Michael Glinsky (CEO Science Leader, CSIRO; Adjunct Professor of](https://reader033.vdocument.in/reader033/viewer/2022060209/5f04411f7e708231d40d11de/html5/thumbnails/25.jpg)
500
1000
1500
2000
2500
3000
3500
0..40..61
1..62..54
6..310
15.825.139.163.1100
158.5
500
1000
1500
2000
2500
3000
3500
500
1000
1500
2000
2500
3000
3500
CMP0
CMP1
CMP2
CMP3
CMP4
CMP5
CMP6
CMP7
CMP8
CMP9
CMP1
0CM
P11
CMP1
2CM
P13
CMP1
4CM
P15
CMP1
6CM
P17
CMP0
CMP1
CMP2
CMP3
CMP4
CMP5
CMP6
CMP7
CMP8
CMP9
CMP1
0CM
P11
CMP1
2CM
P13
CMP1
4CM
P15
CMP1
6CM
P17
CMP0
CMP1
CMP2
CMP3
CMP4
CMP5
CMP6
CMP7
CMP8
CMP9
CMP1
0CM
P11
CMP1
2CM
P13
CMP1
4CM
P15
CMP1
6CM
P17
0..40..61
1..62..54
6.310
15.825.139.863.1 100
0 100 200 300 4000
1
2
3
4
5
infer
red
resis
tivity
-thick
ness
prod
uct (
104 O
hm.m
)
0 100 200 300 4000
100
200
300
400
500
thickness(m),location
thickness(m),location
infer
red
thick
ness
(m)
Dept
h(m)
Dept
h(m)
Dept
h(m)
0..40..61
1..62..54
6..310
15.825.139.163.1100
158.5
A
A B
C DRe
sistiv
ity (O
hm.m
) co
lorba
r
Wedge model
model smoothing
log grid, no smoothing
examined in detail next 25Wednesday, 22 June 2011
![Page 26: Metropolis methods applied to Bayesian geologic inversionsMetropolis methods applied to Bayesian geologic inversions Michael Glinsky (CEO Science Leader, CSIRO; Adjunct Professor of](https://reader033.vdocument.in/reader033/viewer/2022060209/5f04411f7e708231d40d11de/html5/thumbnails/26.jpg)
50 100 150 200 250 300
960
980
1000
1020
1040
0 50 100 150 200 250 3000.00
0.01
0.02
0.03
0.04
0.05
0.06
940 960 980 1000 1020 1040 10600.00
0.05
0.10
0.15
0.20
Depth(m)
Dep
th(m
)
Prob
abili
ty (p
df)
Prob
abili
ty (p
df)
Thickness(m)Thickness(m)
truth case
truth case
MAP estimate MAP estimate
Thickness and depth resolution
26Wednesday, 22 June 2011
![Page 27: Metropolis methods applied to Bayesian geologic inversionsMetropolis methods applied to Bayesian geologic inversions Michael Glinsky (CEO Science Leader, CSIRO; Adjunct Professor of](https://reader033.vdocument.in/reader033/viewer/2022060209/5f04411f7e708231d40d11de/html5/thumbnails/27.jpg)
1000 2000 3000 4000 5000 6000 7000 8000 9000 10000 11000 12000
1E!15
1E!14
1E!13
1E!12
1E!11
Transmitter-receiver offset(m)
Inlin
e |E
| (V/
Am2)
0.25Hz
0.75Hz1.25Hz
0 1000 2000 3000 4000 5000!0.5
0.0
0.5
1.0
1.5
2.0
2.5
3.0
depth(m)
log10
(rho[
Ohm.
m])
Bayesian smoothing on a logarithmic grid
“Truth case”
Bayesian splitting (simple version)
Bayesian splitting and merging
GA
P
“Bird” model
27Wednesday, 22 June 2011
![Page 28: Metropolis methods applied to Bayesian geologic inversionsMetropolis methods applied to Bayesian geologic inversions Michael Glinsky (CEO Science Leader, CSIRO; Adjunct Professor of](https://reader033.vdocument.in/reader033/viewer/2022060209/5f04411f7e708231d40d11de/html5/thumbnails/28.jpg)
Uncertainty sampling methods for “bird” model
• Local covariances pretty hopeless• Sampling methods only possibility
• MCMC : OK but very demanding (narrow twisty objective)• Bayesianized parametric bootstrap: approximate method which is fairly
good, uses optimization methods very heavilyLo
g10(
rho)
depth
P84
P16
P50model
0 4.5 km
0
2
1
3
28Wednesday, 22 June 2011
![Page 29: Metropolis methods applied to Bayesian geologic inversionsMetropolis methods applied to Bayesian geologic inversions Michael Glinsky (CEO Science Leader, CSIRO; Adjunct Professor of](https://reader033.vdocument.in/reader033/viewer/2022060209/5f04411f7e708231d40d11de/html5/thumbnails/29.jpg)
Log resistivity profiles for individual realizations
3.15
-0.1
log1
0 re
sist
ivity
(ohm
*m)
dept
h (m
)
realization29Wednesday, 22 June 2011
![Page 30: Metropolis methods applied to Bayesian geologic inversionsMetropolis methods applied to Bayesian geologic inversions Michael Glinsky (CEO Science Leader, CSIRO; Adjunct Professor of](https://reader033.vdocument.in/reader033/viewer/2022060209/5f04411f7e708231d40d11de/html5/thumbnails/30.jpg)
Cluster separation to discriminate background from anomaly
back
grou
ndan
omal
y
30Wednesday, 22 June 2011
![Page 31: Metropolis methods applied to Bayesian geologic inversionsMetropolis methods applied to Bayesian geologic inversions Michael Glinsky (CEO Science Leader, CSIRO; Adjunct Professor of](https://reader033.vdocument.in/reader033/viewer/2022060209/5f04411f7e708231d40d11de/html5/thumbnails/31.jpg)
Individual anomaly cluster distributions
P84 = 4.9 ohm*mP50 = 6.6 ohm*mP16 = 7.3 ohm*mmodel = 6.3 ohm*m
P84 = 1294 mP50 = 1388 mP16 = 1472 mstddev = 89 mmodel = 1535 m
logResistivity (ohm*m)
P84 = 2685 mP50 = 2838 mP16 = 3109 mstddev = 212 mmodel = 2787 mP84 = 40 ohm*m
P50 = 54 ohm*mP16 = 69 ohm*mmodel = 40 ohm*m
depth (m)
31Wednesday, 22 June 2011
![Page 32: Metropolis methods applied to Bayesian geologic inversionsMetropolis methods applied to Bayesian geologic inversions Michael Glinsky (CEO Science Leader, CSIRO; Adjunct Professor of](https://reader033.vdocument.in/reader033/viewer/2022060209/5f04411f7e708231d40d11de/html5/thumbnails/32.jpg)
Summary of individual anomaly clusters and background
32Wednesday, 22 June 2011
![Page 33: Metropolis methods applied to Bayesian geologic inversionsMetropolis methods applied to Bayesian geologic inversions Michael Glinsky (CEO Science Leader, CSIRO; Adjunct Professor of](https://reader033.vdocument.in/reader033/viewer/2022060209/5f04411f7e708231d40d11de/html5/thumbnails/33.jpg)
Classification of individual realizationsde
pth
(m)
realization33Wednesday, 22 June 2011
![Page 34: Metropolis methods applied to Bayesian geologic inversionsMetropolis methods applied to Bayesian geologic inversions Michael Glinsky (CEO Science Leader, CSIRO; Adjunct Professor of](https://reader033.vdocument.in/reader033/viewer/2022060209/5f04411f7e708231d40d11de/html5/thumbnails/34.jpg)
Corrections for systematic noise
data decimationnoise model fit
0.51.0
2.0
1/
34Wednesday, 22 June 2011
![Page 35: Metropolis methods applied to Bayesian geologic inversionsMetropolis methods applied to Bayesian geologic inversions Michael Glinsky (CEO Science Leader, CSIRO; Adjunct Professor of](https://reader033.vdocument.in/reader033/viewer/2022060209/5f04411f7e708231d40d11de/html5/thumbnails/35.jpg)
Past track record & future
• Peer reviewed technology for probabilistic seismic and CSEM inversions
• DELIVERY, http://tinyurl.com/ydqk7nc and http://tinyurl.com/ye2ld4f• deliveryCSEM, http://tinyurl.com/yl79g3g
• Seismic inversion well benchmarked on synthetic models and applied to over 25 assets by multiple companies
• results cross validated to wells• pre-drill predictions verified by outcomes of drilling
• CSEM inversion well benchmarked on synthetic models and applied to several real datasets including the Cerah prospect in Block N (Sabah)
• Future• refinement and application of 1D method• extension into 2.5D using finite element methods with automatic grid
refinement• development of simultaneous CSEM and seismic timelapse inversion
35Wednesday, 22 June 2011
![Page 36: Metropolis methods applied to Bayesian geologic inversionsMetropolis methods applied to Bayesian geologic inversions Michael Glinsky (CEO Science Leader, CSIRO; Adjunct Professor of](https://reader033.vdocument.in/reader033/viewer/2022060209/5f04411f7e708231d40d11de/html5/thumbnails/36.jpg)
P.S.: What I am really working on
• Understanding and prediction of self organisation of geologic sedimentation
• Developing and using a new renormalization theory, with linear convergent Wick expansion in terms of complexity of interaction
• Next seminar “Invariant actions -- dynamic DNA”
36Wednesday, 22 June 2011
![Page 37: Metropolis methods applied to Bayesian geologic inversionsMetropolis methods applied to Bayesian geologic inversions Michael Glinsky (CEO Science Leader, CSIRO; Adjunct Professor of](https://reader033.vdocument.in/reader033/viewer/2022060209/5f04411f7e708231d40d11de/html5/thumbnails/37.jpg)
Collaborations and personnel into the future
• OCE Science Leader• Michael Glinsky (planned 2 month residence at Santa Fe Institute, Pawsey Supercomputer
Centre Steering Committee and leading UQ for Resources Grand Challenge)• OCE postdocs
• Karen Livesey, theoretical physicist from UWA• Bela Nagy, mathematics from UBC and Santa Fe Institute
• PhD students• Zac Borden, computer simulation from UCSB• Youssef Mroueh, mathematics from L’Ecole Polytechnique
• Honours student• from UWA associated with my Adjunct Professorship in Physics
• Visitors & collaborators• Moshe Strauss, theoretical physicist from Israeli National Lab• Vivek Sarkar, computational science from Rice University• Henry Abarbanel, theoretical physicist from UCSD• Stephane Mallat, mathematics from L’Ecole Polytechnique• Tarabay Antoun, computer simulation from Lawrence Livermore National Lab
• Industrial application• Chevron SEED project• Woodside 10/11 budgeted technology project
37Wednesday, 22 June 2011
![Page 38: Metropolis methods applied to Bayesian geologic inversionsMetropolis methods applied to Bayesian geologic inversions Michael Glinsky (CEO Science Leader, CSIRO; Adjunct Professor of](https://reader033.vdocument.in/reader033/viewer/2022060209/5f04411f7e708231d40d11de/html5/thumbnails/38.jpg)
Thank you
Earth Science and Resource EngineeringMichael GlinskyCEO Science Leader
Phone: +61 458 196 079Email: [email protected]: www.qitech.biz
Contact UsPhone: 1300 363 400 or +61 3 9545 2176
Email: [email protected] Web: www.csiro.au
38Wednesday, 22 June 2011